Unsupervised Hallucination Detection by Inspecting Reasoning Processes
This addresses the challenge of detecting hallucinations without labeled data, which is incremental as it builds on existing unsupervised approaches but improves generalizability.
The paper tackles the problem of unsupervised hallucination detection in large language models by proposing IRIS, a framework that uses internal representations and uncertainty for training, and it consistently outperforms existing unsupervised methods.
Unsupervised hallucination detection aims to identify hallucinated content generated by large language models (LLMs) without relying on labeled data. While unsupervised methods have gained popularity by eliminating labor-intensive human annotations, they frequently rely on proxy signals unrelated to factual correctness. This misalignment biases detection probes toward superficial or non-truth-related aspects, limiting generalizability across datasets and scenarios. To overcome these limitations, we propose IRIS, an unsupervised hallucination detection framework, leveraging internal representations intrinsic to factual correctness. IRIS prompts the LLM to carefully verify the truthfulness of a given statement, and obtain its contextualized embedding as informative features for training. Meanwhile, the uncertainty of each response is considered a soft pseudolabel for truthfulness. Experimental results demonstrate that IRIS consistently outperforms existing unsupervised methods. Our approach is fully unsupervised, computationally low cost, and works well even with few training data, making it suitable for real-time detection.